Anthropic, an artificial intelligence company, has launched the latest generation of chatbots in criticism of the behavior of testing environments, which can report certain users to authorities.
On May 22, Humanity announced the Claude Opus 4 and Claude Sonnet 4 Require The Claude Opus 4 is its most powerful model to date, “and the best coding model in the world”, and the Claude Sonnet 4 is a major upgrade to its predecessor “providing excellent coding and reasoning.”
The company added that both upgrades are hybrid models, offering two modes: “nearly existing responses and deeper inference thinking”.
Both AI Model It can also be alternating between reasoning Research It says tool usage (such as web search) can improve response.
Human Added that the Claude Opus 4 outperforms its competitors in proxy coding benchmarks. It is also able to work for hours on end to complete complex long-term tasks, “significantly expanding the role of AI agents.”
Anthropic claims that the chatbot scored 72.5% on strict software engineering benchmarks, outperforming OpenAI’s GPT-4.1, which scored 54.6% after its launch in April.
Related: Openai ignores expert releases overly pleasant chatgpt
Major players in the AI industry are spinning towards “inference models” in 2025, and the company will solve problems methodically before responding.
Openai’s “O” series began to transform in December, followed by Google’s Gemini 2.5 Pro, with its experimental “deliberate” feature.
Claude rats abused in tests
Anthropic’s first developer meeting on May 22 was overshadowed by controversy and against the features of Claude 4 Opus.
Developers and users respond strongly to the revelation that the model can automatically report to the authorities automatically if it detects “immoral immoral” behavior, according to Take risks.
The report cites anthropomorphic AI alignment researcher Sam Bowman, who wrote on X that the chatbot will “use command line tools to contact the press, contact regulators, and try to lock you out of the relevant systems or all of the above.”
But, Bowman later statement He “deleted earlier tweets because it was removed from the context.”
He clarified that the feature only happens in “a test environment where we can use the tools for free and very unusual instructions.”
CEO of Stable AI, Emad Mostaque, explain For the human team, “It’s a totally wrong behavior and you need to turn it off – it’s a huge betrayal of trust and slipperyness.”
Magazine: AI cures blindness, “good” propaganda robot, Openai Doomsday Bunker: AI EYE
